Artificial neural network-based roughness prediction models for gravel roads considering land use

The Wyoming Technology Transfer Center (WYT2/LTAP) is in the process of developing a holistic and integrated Gravel Road Management System (GRMS). As a part of this effort, gravel roads condition prediction models have been developed. The developed models aim to predict the roughness of gravel roads in terms of International Roughness Index (IRI) considering the land use and the traffic volume levels. Approximately 700 miles of gravel roads in Laramie County, Wyoming were tested and evaluated. The land use of each road was classified into three groups: residential, agricultural, and industrial. While the traffic volume, for the purpose of this study, was categorized into three levels: low, moderate, and high. Both land use and traffic volume levels were found to have significant effect on the IRI deterioration over time. Initially, multiple linear regression was used to develop IRI prediction models. The developed multiple linear regression models represented the data with accepted coefficient of determination (R2) level. Also, Artificial Neural Network (ANN) was used to develop prediction models for the IRI. ANN process started by using 70% of the data to train the model, then the remaining 30% were used for validation and testing. One ANN prediction model was developed for each land use. ANN prediction models showed significantly more ability to represent the data than multiple linear regression. The developed prediction models can be an essential part for developing a holistic GRMS and improving better understanding for gravel roads deterioration behavior.